Asia–Pacific Economic Statistics Week 2019

Integrating economic statistics in monitoring the Agenda 2030

17-21 June 2019 | Bangkok, Thailand

Asia-Pacific Economic Statistics Week Seminar Component

Bangkok, 17-21 June 2019

Name of author Amanda Pratama Putra, Ignatius Aditya Setyadi, Siim Esko, Titi Kanti Lestari Organization BPS – Statistics Indonesia

Positium LBS Contact address BPS - Statistics Indonesia Jl. Dr. Sutomo No. 6-8 Jakarta 10710 Indonesia Contact phone +62878 6396 9334 (Putra),

Email amanda.putra@bps.go.id, adityasetyadi@bps.go.id, siim.esko@positium.com, titi@bps.go.id

Title of Paper

Measuring Commuting Statistics in Indonesia Using Mobile Positioning Data

Abstract

Commuting patterns refers to the movement of workers from their homes (origin) to

work areas located in different cities. Commuter movements generally take place in

metropolitan cities as centers of economic activity. Accurate and up-to-date data about

commuting patterns needs to be obtained continuously as a basis for determining policies

related to rapid urbanization and sustainable urban transportation.

The Population Census and surveys are the traditional methods to used to commuter

statistics. These methods could provide an amount of information concerning commuting flow

and the characteristics of workers grouped by their origin and destination. Indeed, BPS-

Statistics Indonesia has conducted commuter surveys in several major cities in Indonesia

since 2014. However, implementation of the survey could not be carried out simultaneously in

all areas due to the limited costs. As a result, the commuter statistics could not be captured

comprehensively. In addition, commuter statistics data from the survey cannot be updated

regularly in a short period of time because it is expensive; and requires a long time to collect

the data from the field.

1

In order to address the weaknesses of traditional methods, this paper utilized the

availability of mobile positioning data as a complementary technique. From our experiment,

the results show that commuting patterns in Bandung City measured using mobile positioning

data provide the same commuting pattern as the survey results. Therefore, it seems mobile

positioning data could be used as an alternative approach to measuring commuting statistics

in all major cities in Indonesia simultaneously.

2

I. Contents

I. Contents 2

II. Introduction 3

III. Data and Methodology 4

Mobile positioning data 4

Home and work identification 5

Commuter identification 6

Household commuter survey data 7

Calibration with population projection 7

IV. Result 9

Result Comparison: MPD vs Traditional Survey 9

V. Conclusion 12

VI. References 13

3

II. Introduction

For citizens who live in major cities in Indonesia, the main places of activity, such as

work or school that are administratively different from places of residence, are a common

phenomenon today. More facilities offered in the city center the high price of land and building

(including its tax) are the main attraction for residents living in the surrounding (suburbs) area.

In addition, the high cost of housing in the city center makes residents prefer living in satellite

areas that are close to the city center. They travel back and forth from home almost every day

to work. This routine then forms a travel pattern, namely commuting patterns.

Commuting patterns refers to the daily movement of people especially workers from

their homes (origin) to work area. The existence of commuters has a positive impact on the

city center, which can reduce the density of the population living in the cities and the growing

development of satellite areas as a place for commuters. However, the existence of

commuters is a big challenge for big cities to improve public facilities, especially in the

transportation sector that directly affect citizens daily lives. This relate to the SDGs Indicator

11 (Make Cities and Human Settlements Inclusive, Safe, Resilient and Sustainable). Target

11.1, by 2030, ensure access for all to adequate, safe and affordable housing and basic

services and upgrade slums. Target 11.2: By 2030, provide access to safe, affordable,

accessible and sustainable transport systems for all, improving road safety, notably by

expanding public transport, with special attention to the needs of those in vulnerable situations,

women, children, persons with disabilities and older persons.

Commuter movements generally take place in metropolitan cities as the center of

economic activity. Since 2014, BPS – Statistics Indonesia has conducted the commuter survey

in some metropolitan areas in Indonesia to produce statistics and monitoring systems that are

able to provide input for the formulation of population mobility policies and the development of

origin and commuter destinations. The conventional methods like survey could provide the

amount of commuting flow information and the characteristics of workers grouped by their

origin and destination. However, one of the survey-based drawbacks is it takes a long time

from the design of a survey to the release of its final result. Besides, the expensive survey

costs make the implementation of commuter surveys in Indonesia unable to be carried out

simultaneously in all metropolitan areas.

The accurate and up-to-date data about commuting patterns is very challenging to be

obtained continuously as a basis for determining policies related to rapid urbanization and

sustainable urban transportation. Consequently, different approaches and data sources are

needed to measure the commuting statistics more intensively and comprehensively.

Traditional census and administrative data are solid sources for long-term processes. For the

4

short term and everyday mobility, more flexible methods such as crowdsourcing, satellite-

based methods, or modern sensing technologies are needed.

To reduce the time lags and cost in producing the commuting statistics, there are a lot

of attempts related to experimental work to use other data sources. Novak et al. [1] in 2013

explore the potential use of mobile phone location data to map the commuting patterns in

Estonia. They found that mobile phone location data represent a significant alternative source

in cases where traditional data are missing or are not appropriate or the method of collection

is not cost effective. It implies that the commuting behaviors from big data (mobile positioning

data) can allow the public sector to produce more frequent and more detail statistics.

As part of innovation and exploiting the new sources of data, we are currently working

with alternative data sources for commuting statistics using mobile positioning data (MPD).

MPD is the pinpoint location of mobile network subscribers which extracted from the location

of exposed base transceiver station (BTS). The data is obtained from Telkomsel, one of the

largest mobile network operator (MNO) in Indonesia. Since 2017 BPS has begun to

collaborate with Telkomsel to explore MPD as an alternative and complementary data source

for some official statistics production. The data obtained is based on contract for a year. While

for the methodology for commuting we got help from Positium LBS. So, the methodology that

we use is similar as in Novak et. al paper. This paper aims to describe the result of the

experience using MPD to detect the commuting pattern, that will be used as the basis for

measuring commuting statistics. Finally, we evaluate the result with official commuting

statistics based on the survey result.

The paper is organized as follows. Section 3, describes the data and methodology.

Section 4, provides the results. The last section is the conclusion.

III. Data and Methodology

Mobile positioning data

Mobile positioning data in this paper refers to the large-scale location data of

subscribers of MNOs that are processed and stored in operator systems (Altin et al. 2015).

The location of the mobile devices can be obtained in real-time or historically. The data is

highly sensitive but also very valuable. It could be used anonymously and aggregated in the

high-level aggregation to ensure customer privacy. MNOs store huge amounts of data for their

business purposes including “location events” (e.g. network infrastructure quality monitoring

(signaling quality), operational activity data, billing data, data for marketing and CRM,

handover, Location Area updates, active positioning requests, etc.) (Tiru, 2014). In Indonesia,

call detail records (CDR) and signaling data are used since Telkomsel is collecting, storing,

and commercializing it. The data period used for measuring the commuting pattern are during

5

2018. Both CDR and signaling data have time and location information (a detailed latitude and

longitude coordinate for cell towers) for every detected event. Telkomsel processes the CDR

and signaling data into staypoint based on Mobility Genome Framework. Staypoint is data

points that have been grouped to the constraint of average distance and the total duration of

stay. according to Dang, et. al. (2017), “a stay point is constructed if the average distance from

its centroid to all other points is less than ∆d and total duration spent is more than ∆t”.

Owing to the privacy policy, there are several restrictions in the cooperation agreement

with Telkomsel including data access rights. Telkomsel applies a sandbox mechanism

approach in the process of developing algorithms and accessing data to protect data

confidentiality. The data provided is staypoint (processed by Telkomsel) due to access to raw

data is restricted. Then the algorithm is set to work on staypoint data. Because of the limitation,

we were unable to estimate in the low administrative level (Desa/Kelurahan or even

Kecamatan), so we decided to use the Kabupaten level which this level also adequate to

compare with official statistics.

Home and work identification

Figure 1. Usual Environment (Saluveer, E)

Implementing home and work to mobile positioning data is quite a tricky part.

Therefore, BPS-Statistics Indonesia asked about 63 employees in all over Indonesia if they

like to be volunteers. BPS uses the MPD of volunteers to develop an algorithm of usual

6

environment identification, especially home and work and then validate the result for each

individual with their actual home and work location at the municipal level. At first, because the

home and work location for each volunteer is estimated based on one past month data only,

the results become inaccurate (only 58,92% accurate). Then, home and work location are

estimated based on past 6 months data which showed more accurate results (up to 90%

accurate) compared to home and work locations that were estimated based on 1-month data.

The long-term home and work location decided for the past 6 months and for every

month select highest ranking home and work (based on duration). Then sum durations of

home and work candidates by grouping them by Kabupaten (municipality) and Kabupaten with

the highest duration will be the actual location for home/work.

Commuter identification

In this paper, commuters are defined as subscribers who live and work in different

municipal and make a trip from home to work and return home within 24 hours routinely.

Currently, it is determined that for the observation month, a subscriber will be classified as a

commuter when he has at least 4 days during that month he has made a trip that follows the

aforementioned rule. These 4 days also have to be registered on 4 different weeks, meaning

that if a subscriber travels between home and work municipal on daily basis for the first two

weeks and then the next two weeks are on a vacation, they will not be counted as a commuter.

Commuters with very distant locations (no direct transportation connections) are

removed (outliers). For example, if we have just couple of subscribers commuting from Jakarta

to distant kabupaten X, then those commuters will be removed. Commuters with non-standard

work patterns, such as night workers might not be identified by this approach. Additionally,

commuters who are on holiday, ill or otherwise absent from work in some specific month may

not be identified because we process the algorithm monthly. Figure 2 describe the Commuters

algorithm proposed by BPS and implemented by Telkomsel.

Figure 2. Commuters algorithm

7

Household commuter survey data

The latest commuter survey conducted by BPS-Statistics Indonesia was held in 2017,

with domain estimation covering two different areas, the (1) Bandung Raya and (2)

Gerbangkertosusila.

Figure 3. Area of Bandung Raya and Gerbangkertosusila

This survey aims to understand the behavior of commuters in 5 cities or regencies in

Bandung Raya (Kabupaten Bandung, Kabupaten Sumedang, Kabupaten Bandung Barat,

Kota Bandung, and Kota Cimahi) and 7 cities or regencies in Gerbangkertosusila (Kabupaten

Gresik, Kabupaten Bangkalan, Kabupaten Mojokerto, Kota Mojokerto, Kota Surabaya,

Kabupaten Sidoarjo, and Kabupaten Lamongan). The commuter survey is a household survey

covering 10.600 households from 1.060 enumeration areas. the time reference used in this

survey is June 2017. The survey provide the estimation of flows in the areas and also provide

the demographic characteristics of the commuters as well.

Calibration with population projection

BPS has the duty to produce the latest population data and information, and its

estimates in the future as a basis for determining development policies. Population projection

is a scientific calculation based on the assumptions of the components of the population

growth rate, namely birth, death, and displacement (migration). These three components will

determine the number and age structure of the population in the future. The population

includes all residents in the geographical area of the Republic of Indonesia who has stayed

8

for six months or longer and those who intended to stay even though their length of stay was

less than six months.

To validate our usual environment algorithm, especially home identification in

aggregated level, we compare the number of subscribers in a municipal area with the

population projection for all 514 regions. MPD’s home was only obtained from one MNO (only

Telkomsel), then we have to calibrate those home with the population projection data in each

municipality, so that can reasonably be compared with the statistics result from the survey.

The result is shown in Table 1. For example, the calibration ratio for Kota Bandung is 1.44,

Kota Surabaya is 1.22, and Kabupaten Sumedang with 1.04.

Table 1. Calibration rate between Home from MPD and Population Projection

No BPS Population Projection BPS Population Projection Home for Distinct

Telkomsel Number Ratio Population Projection

and Home

(1) (2) (3) (4) (5)

1 Bandung 3,697,346 1,510,948 2.44

2 Sumedang 1,148,766 657,616 1.74

3 Bandung Barat 1,677,985 518,348 3.24

4 Kota Bandung 2,501,823 1,736,766 1.44

5 Kota Cimahi 605,575 372,089 1.62

6 Gresik 1,294,343 652,830 1.98

7 Bangkalan 976,223 329,405 2.96

8 Mojokerto 1,105,643 323,267 3.42

9 Kota Mojokerto 127,948 57,635 2.22

10 Kota Surabaya 2,881,942 2,350,916 1.22

11 Sidoarjo 2,205,717 1,213,393 1.82

12 Lamongan 1,188,773 437,846 2.72

9

IV. Result

Result Comparison: MPD vs Traditional Survey

For this study, we only use the number of commuters who have a home location and

work location within the area that we just mentioned. So, the number of commuters used in

this study (100%) does not take the number of commuters who have work locations outside

the area mentioned. Furthermore, in table 1 and table 2, the number of commuters is

estimated according to the location of the home and work respectively. In the table, two

estimation numbers are obtained from both, MPD and traditional surveys. the numbers in

brackets and thickened are estimated numbers from MPD. The commuting patterns in

Bandung Raya visually presented in the figure 4.

Figure 4. The Commuting Pattern in Bandung Raya

Table 2 is the origin destination matrix (ODM) for commuters in Bandung Raya to

shows the result comparison of MPD and traditional survey. For instance, in the total estimated

number of commuters in Bandung Raya, the traditional survey estimate 36.05% of people

commute from Kabupaten Bandung to Kota Bandung, while MPD estimate for 43.22%. The

result also shows that Kota Bandung is the most favorite destination for commuters in every

Bandung Raya origin, except commuters from Kabupaten Sumedang that only 2.88%

commute to Kota Bandung.

10

Table 2. Origin Destination Matrix (ODM) of Commuters in Bandung Raya

Home (Origin)

Work (Destination)

Bandung Bandung

Barat Kota

Bandung Kota

Cimahi Sumedang

Bandung - 0.89%

(1.21%) 36.05%

(43.22%) 3.31%

(1.99%) 5.72%

(3.41%)

Bandung Barat

1.12% (1.53%)

- 10.00% (15.92%)

8.98% (6.20%)

0.00% (0.23%)

Kota Bandung

6.53% (6.55%)

1.78% (1.74%)

- 4.48% (2.56%)

0.56% (0.87%)

Kota Cimahi

0.89% (0.80%)

1.88% (1.66%)

10.61% (8.31%)

- 0.12% (0.06%)

Sumedang 4.11%

(1.83%) 0.00%

(0.09%) 2.88%

(1.78%) 0.09%

(0.04%) -

Table 3 is the ODM of commuters in Gerbangkertosusila. Following the result, both

MPD and traditional survey estimate Kota Surabaya as the top destination for commuters. On

the other hand, Kota Mojokerto that also labeled as a city like Kota Surabaya is only favorite

for the commuters from Kabupaten Mojokerto.

Table 3. Origin Destination Matrix (ODM) of Commuters in Gerbangkertosusila

Home

(Origin)

Work (Destination)

Gresik Bangkalan Mojokerto Kota

Mojokerto

Kota

Surabaya Sidoarjo Lamongan

Gresik - 0.00%

(0.20%)

0.83%

(0.31)%

0.00%

(0.02%)

10.89%

(13.28%)

2.17%

(1.35%)

2.27%

(2.05%)

Bangkalan 0.00%

(0.30%)

- 0.00%

(0.00%)

0.00%

(0.00%)

1.23%

(2.58%)

0.00%

(0.08%)

0.00%

(0.00%)

Mojokerto 1.62%

(1.12%)

0.10%

(0.02%)

- 8.65%

(3.34%)

2.34%

(3.01%)

4.99%

(3.25%)

0.29%

(0.54%)

Kota

Mojokerto

0.05%

(0.07%)

0.00%

(0.00%)

2.09%

(1.04%)

- 0.42%

(0.40%)

0.76%

(0.35%)

0.00%

(0.01%)

Kota

Surabaya

5.07%

(4.41%)

0.32%

(0.52%)

0.37%

(0.39%)

0.00%

(0.08%)

- 9.86%

(8.79%)

0.00%

(0.21%)

Sidoarjo 5.25%

(2.43%)

0.00%

(0.12%)

3.89%

(1.44%)

0.65%

(0.27%)

29.00%

(42.27%)

- 0.00%

(0.11%)

11

Lamongan 4.77%

(3.48%)

0.00%

(0.01%)

0.00%

(0.25%)

0.00%

(0.02%)

2.03%

(1.73%)

0.09%

(0.15%)

-

Figure 5 Commuting Pattern from Traditional Survey

The cities with more facilities offered like Kota Bandung and Kota Surabaya are the

main places for commuters to work and make a living. It makes sense because the two cities

are the center of provincial government and they are located right in the middle of their area.

Figure 5 shows the commuting pattern from the traditional survey and figure 6 shows the

commuting pattern from MPD. In the diagrams, each flow visualizes a specific flow between

origin to destination. It describes clearly that these two different methods relatively produce

the same flow.

12

Figure 6 Commuting Pattern from MPD

Mathematically, we provide the calculation of Pearson Correlation Coefficient (PCC)

indicates that there is a high positive association between the result from official statistics and

MPD. The PCC for the estimation in Bandung Raya is 0.98 (strong correlation) and for

Gerbangkertosusila is 0.96 (strong correlation).

Figure 7 Linear correlation

V. Conclusion

The use of mobile positioning data changes the process of design, build, data

collection, data processing, and dissemination. Data collection will be more efficient in terms

of cost and time. Near real-time dissemination can be achieved by using mobile positioning

data. Big data is a part of the data revolution and needs to develop as a validated source for

statistics.

The results showed that mobile positioning data has the potential to increase the level

of disaggregation, timeliness, and usefulness for commuting analysis. The method for

calculating commuters is using mobile positioning data and calibrating it with population

projection data and also comparing with Commuter Household Survey for benchmarking. The

commuter household survey can only be conducted every two years and only covered some

municipalities. So, the data frequency is quite low and the data quality of household survey is

depending on respondent’s answer, while Mobile Positioning Data can provide commuter data

from all the municipalities (514 Kabupaten in Indonesia), more frequent data and also it

depends on methodology that can be improved over time and data recalculated as the

13

methodology improving. According to the limitation of our current methodology for MPD that

only get data source from single mobile networks and not be able to detect commuters with

non-standard work patterns, we believe that our approach is still under-coverage even after

we calibrate the result with the population projection.

VI. References

[1]. Ahas, Rein. Siiri Silm, Erki Saluveer, and Olle Jaarv, (2009). Modelling Home and Work

Locations of Population Using Passive Mobile Positioning Data,

https://www.researchgate.net/publication/227086074.

[2] Ahas, Rein. Siiri Silm, Olle Jaarv, Erki Saluveer, and Margus Tiru (2013). Using Mobile Positioning Data to Model Locations Meaningful to User of Mobile Phones, Journal of Urban Technology, 17:1, 3-27.

[3] Altin, Laura. Tiru, Margus. Saluveer, Erki. Puura, Anniki. (2015). Using Passive Mobile Positioning Data in Tourism and Population Statistics, the New Techniques and Technologies for Statistics 2015 proceedings. Brussel: Eurostat.

[4] Anh Dang, The & Deepak, Jayakumaran & Wang, Jingxuan & Luo, Shixin & Jin, Yunye & Ng, Yibin & Lim, Aloysius & Li, Ying. (2017). Mobility Genome™- A Framework for Mobility Intelligence from Large-Scale Spatio-Temporal Data. 449-458. 10.1109/DSAA.2017.20.

[5] Novak, Jakub. Ahas, Rein. Aasa, Anto. and Silm, Siiri. (2013). Application of mobile phone location data in mapping of commuting patterns and functional regionalization: a pilot study of Estonia. Journal of Maps, 9:1, 10:15, DOI: 10.1080/17445647.2012.762331.

[6] Raun, Janika and Rein Ahas. (2016). Defining usual environment with mobile tracking data

, the 14th Global Forum on Tourism Statistics, Venice, Italy.

[7] Tiru, Margus. (2014). Overview of the Sources and Challenges of Mobile Positioning Data for Statistics, presented in the International Conference on Big Data for Official Statistics. Beijing: UN Trade Statistics, Positium, and University of Tartu.